Main Notebook in Detail
how deep should we go?
lets discuss that regarding time
(presentation should be 10 minutes, followed by 5 minutes of Q&A)
Deep Learning Tutorial
problem setting (policy relevance)
dataset geneva
model & methods
prototypical Networks
main notebook in detail
results to expect
wrap-up
Tutorial Task: Why rooftop segmentation matters (solar planning, urban policy)
Cities need rooftop maps for solar PV planning
Manual labeling is expensive
Few-shot learning reduces annotation cost
Geneva dataset comes in (example for cities)
Satellite Images: High-resolution RGB satellite images of Geneva
Switzerland Segmentation Labels: Binary masks indicating rooftop locations
Resolution: Images at various resolutions suitable for few-shot learning
Train/test sizes (val?)
Example image + mask
Overlay visualization (the ones you already generated!)
Challenges: small rooftops, shadows, label noise, class imbalance
Insert & dont forget: visuals notebook data preprocessing etc
Data Preprocessing
Model Architecture
Few-Shot in a Nutshell (modified figure from paper)
Few-Shot in implementation (ntoebook reference/ pseudocode for logic?)
Training strategy
Loss function
Evaluation metrics
Insert modified figure here
high-level schematic (support → prototype → similarity → segmentation)
literature reference: SRPNet
how deep should we go?
lets discuss that regarding time
(presentation should be 10 minutes, followed by 5 minutes of Q&A)
Show performance for 1-shot / 5-shot / full-data comparison
Show predicted masks
Open to Discuss:
strengths
weaknesses
failure cases (shadows, tiny rooftops)
insert more from discussion + memo here
What we have so far:
insert bullet point here
insert bullet point here
What we still need to finalize:
insert bullet point here
insert bullet point here
Questions to discuss in class/ lynn